Steve BARRETT wrote:
>
> We have recently converted our network from a combination of
> bridged point-to-point and multiplexed analog services to a
> routed Frame Relay network.  We use Cisco 2503 routers at 18
> remote sites and a 4500 at the hub site. [snip]...
> We are also considering allocating separate PVCs to support the
> bridging independently of the TCP/IP traffic (Note:
> Approximately 90% of our user equipment is serially
> connected.)

I presume that implies remote DTCs as well as the 2503s, so we can't
go with telnet/NS-VT connections to eliminate bridging...

> With that all said, does anyone with a similar environment
> have statistical or anecdotal information about how much
> bandwith is sufficient for a given number of heavily used
> Internet workstations (i.e., how many Netscape workstations
> can I reasonably expect to support with a 128K PVC)?

Addressing just TCP/IP directly, a 19.2kbps PPP dialup is tolerable for
my home use (other than downloading).  In the office I have 10Mbps
locally (in contrast).  Some other figures - we are on a statewide
network with Internet connections to BBN two hops away.  Our "slice"
of the fractional T1 is 768kbps for our entire campus (1500 nodes now
and growing) with a tail circuit (in effect; actually feeding out the
other side of a ring) of 320kpbs to Jackson and UT-Martin.  The state is
routing TCP/IP and IPX; we're locally routing IP, IPX, and Appletalk.
If you are using IPX or AT, I hope you are routing them; else they will
figure into your bridging load and they aren't broadcast-friendly.  We
also act as an ISP for a local county school system, providing service
over a 56kbps line to ~80 nodes, though judging from traffic I'd have to
say access to students must be restricted :-)

The separate PVC for bridging sounds like a wonderful idea, but I would
try and grab some traffic guesstimates from your current configuration.
Suggest you do 'show interface accounting' on several of those routers
to get a breakdown of packet and byte counts by protocol, paying special
attention to the "bridged" traffic per interface or VC.  Your aggregate
bridge traffic in the central office is probably high, but as long as
the remote ends report smaller numbers, the spanning tree is working and
isn't too large (be sure to reset counters before making comparative
readings).  The only "catch" would be isolating the bridged PVC from
your other TCP traffic.  If you have two LANICs in your 3000, this will
be trivial (one LANIC for DTSLINK/bridge, one for WAN/TCP); otherwise
I'm not sure how you'd manage this as a separate pvc.

Otherwise, assuming you have your remote routers configured with default
routes and not bothering with any chatty routing protocols between them
and the central router, you should pretty much have a clean pipe for
Netscape or whatever other services you want to pull without any other
interfering overhead (it is generally some router overhead that tends to
turn down the bandwidth much faster than simple contention between
equivalent browsers/customers; router updates, IPX SAPs, bridging
updates, etc tend to get priority).

Jeff Kell <[log in to unmask]>